Validation of Brief Condom Use Attitudes Scales For Spanish Speaking People-who-use-drugs in El Salvador

People-who-use-drugs (PWUDs) are a population severely impacted by a concentrated epidemic of HIV. Behavioral interventions to prevent and treat HIV among PWUDs have been implemented around the world including in low- and middle-income countries which have been disproportionately affected by HIV. However, few studies have been conducted with PWUDs to assess the validity and psychometric properties of measures that assess intervention effectiveness. Furthermore, there are very few measures that have been validated on transnational populations as most research interventions have utilized U.S. samples. – – Results indicated that the two-factor, 8-item correlated model for the CUAS-SSF scale had an excellent fit with an adequate reliability (α = .764). The confirmatory factor analysis for the 5-item, CUSN-SSF scale indicated a satisfactory fit with 3 of 6 fit indices indicating adequate fit. Analysis of the two-factor, 5-item CUSE-SSF scale indicated satisfactory fit with an adequate reliability (α = .844). Bivariate analysis indicated correlations between all measures and with self-reported condom use. Results indicated that these brief measures are valid and reliable and can be utilized to assess the effectiveness of HIV risk reduction interventions among Spanish speaking PWUDs.

these four key populations: sex workers, transgender individuals, men-who-have-sex-with-men (MSM), and people-who-use-drugs (PWUDs), and their sexual partners [1]. Specifically, it is estimated that risk for acquiring HIV is 27 times higher for MSM, 23 times higher for people who inject drugs, 13 times higher for female sex workers, and 12 times higher for transgender women than the general population [1]. Furthermore, a great proportion of concentrated epidemics, or sub-epidemics, occur in places such as Latin America among individuals from the abovementioned vulnerable populations [2].
HIV prevalence for the general population in Latin America is approximately 1% [2]. However, high HIV prevalence among many key populations in Latin America is a significant concern. Although surveillance data for sex workers, MSM, and transgender individuals has been consistently collected in Latin America and showed epidemic levels of HIV incidence [1, [3][4][5][6], very little data has been collected on PWUDs. Difficulties in estimating HIV prevalence for PWUDs has been well documented as very few countries collect official surveillance data on drug users [7]. For instance, it was estimated that only Brazil, Chile, Colombia, Mexico, Nicaragua, Paraguay, and Peru had formally collected information on HIV prevalence among PWUDs since 1987 [7]. This is especially problematic as recent surveillance data estimated that illegal drug consumption was approximately 4.8% higher in Latin American countries than the global average of 3.8% [8]. The limited information available indicates that Latin America had approximately 1,508,000 to 2,597,500 injection drug users of which 181,500 to 1,175,500 most likely were HIV positive [7] and Central American countries may bear the brunt of drug use incidence and prevalence. For example, a large study examined illicit drug use across multiple countries in Central America and found that actual use of crack or coca paste for adolescents in El Salvador was the highest at 8% compared to other countries [9]. A national survey around this same time period found that 1% of residents in San Salvador had consumed crack within the last 30 days [10]. However, crack use rates were higher (4.9%) among adolescents who were 18years-old [10].
HIV prevalence for the general population in El Salvador is estimated to be 0.8% [11] [12]. However, multiple sources have indicated alarming, epidemic levels of HIV and other STDs in localized pockets of key populations. HIV and STD rates for MSM has been estimated to be between 7.7% and 15.5% [1, 13,14] between 3.14% and 6% for sex workers [14,15], and between 20.7% and 71.1% for transgender women [4,14,15]. As aforementioned, no official surveillance data reports on HIV/STDs prevalence among individuals who consume drugs in El Salvador exists; however, research studies estimated HIV prevalence in crack users to be between 4.9% to greater than 8% [16]. High levels of sexual risk factors/behaviors among crack users is what makes this population particularly vulnerable to contracting HIV [17] as crack use has been associated with higher numbers of sexual partners [18], increased impulsivity and sex drive [19], more barriers to acquiring healthcare and high levels of sexual assault and violence [20], and exchanging sex for drugs [21,22]. As a result, there is an urgent need to design and implement interventions and assess their effectiveness using appropriately validated outcome measures within relevant cultural contexts.
Condom use is a primary outcome assessed in most HIV prevention interventions with up to 51% of research studies in developing countries measuring condom use as a primary outcome for intervention effectiveness [23]. This is the case because other forms of prevention such as preexposure prophylaxis is not readily available and uptake would need to increase by a greater amount to achieve the same amount of effectiveness as condom use at the population-level [24].
Furthermore, consistent and correct use of condoms is the cheapest and most effective way to reduce HIV/STD transmission [25][26][27][28][29]. Moreover, much theory and subsequent research has hypothesized and found associations between attitudes, norms, and self-efficacy with intentions to use condoms, condom use frequency, and number of sexual partners [30,31]. However, many evidence-based HIV interventions grounded in these theories have been designed and funded by the United States which has led to an implementation gap in other communities possibly due to lack of cultural considerations or political and economic issues [32][33][34][35] to assess intervention efficacy even though low-and middleincome countries bear most of the brunt of HIV and AIDS infections and deaths [36]. It is only recently that research on the effectiveness of HIV prevention interventions and psychometric properties of outcome measures has been conducted in transnational contexts. Consequently, the dearth of psychometric validation studies has led to limitations in assessing the effectiveness of HIV/STD interventions in non-US contexts.
The purpose of the present study is to investigate the construct validity of three brief condom use attitude scales among a large sample of Spanish mono-lingual speaking crack/cocaine users living in San Salvador. The validation of brief measures to assess HIV prevention effectiveness among a population that may exhibit some temporary cognitive interference or decline such as individuals who are actively using drugs is important. To assess construct validity we computed exploratory and confirmatory factor analyses and correlations. To our knowledge no measures currently exist that have been validated on Spanish-speaking PWUDs. This is concerning as Latin America has multiple pockets of key populations that have high rates of HIV/STDs. Because condoms are considered one of the most effective and affordable ways to reduce HIV transmission, it is important that interventions target and reliably assess condom use behaviors and attitudes.

Methods Participants
Participants were 1,324 crack users (87% male; <1% transgender). Ages ranged from 18 to 70 with a mean of approximately 36 years old (SD = 11.337). Approximately 51% of the individuals indicated that they did not have a stable place to live. Average monthly wage was $231 (SD = 213.34) with a range of $2.50 and $1,500. Approximately 71% had less than a 9th grade education. The average number of times within a 30-day period in which participants used crack was approximately 14 times (SD = 23.214). Table 1 presents other demographic characteristics.  items more relevant to PWUDs in a different cultural context. Evaluation from Spanish-speaking experts in the field drove decision-making in which items to alter and retain. Participants were recruited via respondent-driven sampling (RDS) to answer the survey. In RDS, identified individuals from the community, or "seeds," worked in conjunction with community partners to recruit participants [37]. To be eligible to be a seed, the participant had to meet inclusion criteria and be willing to provide a list of individuals in their social network who they believed to be at high-risk for acquiring HIV. These seeds subsequently were given coupons to handout to other potential participants to take part in the study [38]. To be eligible for the study, participants had to be at least 18 years of age, smoked crack in the last month, and be able to provide informed consent. All surveys were administered face to face in a private location. All participants received $5 for participating in the survey and $2 for each participant they recruited. The Medical College of Wisconsin IRB approved the study.

Measures
Demographics. Participants were asked to self-report their age, gender, education, marital status, sexual orientation, income, and homelessness. Participants were also asked about frequency of crack use, number of sexual partners, condom use, number of sexually transmitted diseases ever acquired, and whether they had previously tested for HIV. Frequency of substance use and substance of choice was assessed by asking the participants to review a list of substances and indicate if they had ever used the substance (i.e., yes or no) and how often they had used in the last 30 days. Condom use was assessed by asking participants to respond to how often they had vaginal or anal sex without using a condom in the last 30 days. To assess whether participants had been diagnosed with sexually transmitted diseases, including HIV, participants were given a list of common STDs and asked to respond (i.e., yes or no) if they had ever been diagnosed with any of the STDs previously.
Additionally, the participants were asked how often they had been diagnosed with STDs or the year in which they had been diagnosed with HIV.

Data Analysis
Data analysis were conducted with statistical package for the social sciences (SPSS) v. 25 and Amos v. 25. As the condom use attitudes, self-efficacy, and norm scales were abbreviated, translated, and adapted to be used for PWUD, an exploratory factor analysis (EFA) was first performed to identify a data-driven model of each scale followed by a confirmatory factor analysis (CFA) that was used to validate the structure identified in the exploratory factor analyses. To conduct EFAs and CFAs, we followed recommendations to split the data randomly into two datasets comprising approximately 50% of the cases per dataset [41]. Subsequently, three exploratory and confirmatory factor analyses were conducted separately for each of the scales. The CFA was used to confirm the hypothesized structure based on the initial EFAs. EFAs were conducted using the first half of the randomly split data (N = 643) and the CFAs were conducted using the second half (N = 681).
Prior to any statistical analysis, data were inspected for missing values and multivariate normality.
Missing data was less than 5% which is within an acceptable range [42]. Missing values were replaced by the series mean as simulation studies have demonstrated that this procedure produces similar values to regression and expectation maximization techniques when missing values fall below the 5% threshold [42]. Assumptions of multivariate normality were assessed using kurtosis, skewness, scatterplots, and histograms. Skewness and kurtosis values for each scale at the scale-level fell within the acceptable range of +/-2 [43]. Additionally, skewness and kurtosis values at the item-level fell within the acceptable range of +/-2 in both datasets [43].
In EFAs, sample fit was identified using the Kaiser-Meyer-Olkin method and the Bartlett sphericity test [44]. A KMO value above .70 was considered acceptable [45]. Furthermore, to determine the number of factors in EFAs, we used multiple criteria, which included Kaiser's Rule, inspection of the Scree Plot, interpretability of the pattern of loadings, simple structure, and amount of variance explained [46][47][48][49]. The factor matrix was evaluated to determine the number of items retained on each factor and thereby the interpretability of the factors [47]. The criteria of .32 and a difference of at least .17 between cross-loading items were utilized to determine the item selection for each factor [50]. Lastly, to rotate factors during EFAs, we employed an oblique rotation method (i.e., promax) as it has significant advantages over an orthogonal rotation, and previous literature and theory suggest that factors can be correlated and multidimensional [51].
We employed Maximum Likelihood (ML) estimation for all exploratory factor analysis as data was determined to be normally distributed. ML has the added benefit of yielding a wider range of fit indices [51] and allows for factors to be correlated [52]. (40%) and demonstrated a simple two-factor structure. All items had factor loadings above .32 and had cross loadings that were at least .17 in difference. Items 1, 2, 4, and 6 loaded onto Factor 1 which we labeled "general attitudes towards condom use". Items 3, 5, and 8 all loaded onto Factor 2 which we labeled "attitudes towards condom use with partners". The factor correlation matrix indicated a moderate correlation between the two factors (.625). Cronbach's Alpha indicated an acceptable reliability (α = .764). Table 2 presents the final pattern matrix. .712 For the CFA, we hypothesized a two correlated factor model with four items (i.e., items 1, 2, 4, and 6) loading on the first factor and three items (i.e., items, 3, 5, and 8) loading on the second factor. The model yielded a χ2 (13) = 50.190, p < .001; GFI = .980, CFI = .962, TLI = .939; RMSEA = .065, and SRMR = .0342. Therefore, all of these values met the a priori specified values which indicate adequate fit. Additionally, the results indicated that both Factor 1 (i.e., items, 1, 2, 4, and 6) and Factor 2 (i.e., items 3, 5, and 8) regression weights were all significant (p < .001) with beta values that ranged between β = .39 to .81 and β = .56 to .60, respectively. Factor 1 and Factor 2 were significantly (p < .001) correlated with a covariance weight of .74. The parameter estimates are represented in Fig. 1. (CUSN -SSF). The Kaiser-Meyer-Olkin method (.720) and the Bartlett sphericity test (p < .001) [44] indicated that the overall sample fit and size were satisfactory. Inspection of eigenvalues and the scree plot, and assessment of the interpretability of the pattern loadings and simple structure revealed a two-factor scale which explained approximately 30% of the variance. All items loaded onto the two factors with loadings above .32 and no cross-loadings which is indicative of a simple structure and good interpretability of the pattern loadings. Items 1, 2, 5, 6, and 7 loaded onto Factor 1 while items 3 and 4 loaded onto Factor 2. The factor correlation matrix indicated a small yet meaningful correlation between the two factors (.249). Unfortunately, estimates of reliability indicated that the scale had poor reliability (α = .501). Due to the low reliability of the scale and that one of its factors was composed of two items, a second exploratory factor analysis was conducted without items 3 and 4. Kaiser's Rule, the scree plot, and the interpretability of the pattern loadings indicated only one factor with all five items having above a .32 factor loading and approximately 31% of the variance was explained. Furthermore, reliability for the five-item scale increase to α = .674 which is considered adequate for an exploratory measure [55]. The pattern matrix for the CUSN-SSF is depicted in Table 3. .474 TLI = .832; RMSEA = .099; and SRMR = .0482. Additionally, all items loaded significantly and in the expected direction onto the factor with regression weights were all significant (p < .001) with beta values that ranged between β = .39 to .74. All parameter estimates are depicted in Fig. 2.

Condom Use Social Norms -Spanish Short Form
Condom Use Self-Efficacy -Spanish Short Form. Kaiser's rule and the scree plot suggested one factor which accounted for approximately 80% of the variance. However, the pattern matrix was unable to be interpreted due to the presence of a Heywood case (i.e., item 2). A Heywood case indicates an improper solution which may be the result of highly correlated items or items that have zero or negative variances [56]. Subsequently, we inspected the inter-item correlational matrix to assess for multicollinearity which revealed a very high correlation between items 5 and 6 (r = .88). Simulation studies have examined the impact that multicollinearity has on Type II error and found that these errors can be mitigated by high measure reliability, a large amount of variance, and large sample size [57]. Consequently, item 6 was dropped and a second EFA was conducted. In the second iteration, the KMO method (.723) and the Barlett sphericity test (p < .001) both suggested that the sample fit and size were satisfactory. The Kaiser Rule, scree plot, and pattern matrix all indicated a two-factor structure which explained approximately 80% of the variance. The pattern matrix revealed a simple structure with interpretability of the factor loadings. Items 3, 4, and 5 loaded onto Factor 1 while items 1 and 2 loaded onto Factor 2. The factor correlation matrix indicated a significant correlation between the two factors (α = .399). Lastly, Cronbach's alpha indicated an acceptable reliability (α = .844). Tables 2 and 3 present item-correlations and descriptive statistics (e.g., M, SD). The pattern matrix for the CUSE-SSF is depicted in Table 4. Table 4 Pattern Matrix for the CUSE -SSF items (N = 645) Variable Factor 1 Factor 2 1. Confío en mí, en que hablare con mi pareja sexual acera del VIHA/SIDA aunque este bajo la influencia del crack.
.908 4. Confío en mi en que usare condón la próxima vez que tenga sexo aunque la persona con quien este teniendo sex este en desacuerdo. The purpose of our study was to evaluate the construct validity and psychometric properties of three brief measures of condom use attitudes, social norms, and self-efficacy using exploratory and confirmatory factor analyses and correlational analyses to assess convergent validity. Our results indicate that an 8-item condom attitudes scale and a 5-item condom use self-efficacy scales are brief reliable and valid measures that can be employed to assess factors associated with condom use in Spanish speaking PWUDs. The psychometric properties of non-abbreviated transcultural versions of the UCLA MCAS have been successfully replicated with negligible issues among non-substance users [58][59]. Psychometric validation studies that utilized confirmatory factor analysis also found that the translated measures had a five-factor model using a confirmatory factor analysis and found comparable results (CFI = .935) to the original study (CFI = .90) [60]. In one study, researchers evaluated the structure of the 26-item Echelle Multidimensionnelle des Attitudes Relatives á l'Utilisation du Préservatif (French version of the MCAS) on a sample of 410 female and 354 male undergraduates. The confirmatory factor analyses replicated the five-factor structure in a sample of men and women (CFI = .927 for both) [58]. Overall, these replications indicated that the MCAS continues to be relevant despite possible cultural shifts in sexual attitudes and behaviors. Notably, our analysis of the CUAS -SSF revealed fit index values that were better than both the original UCLA MCAS and the transcultural versions of the scale.
Regarding results for the condom use self-efficacy scale, our findings are less compelling. EFAs revealed a twofactor structure yet low reliability and the CFA yielded inconsistent indices of fit in at least two measures which indicates a moderate fit at best. As post-hoc analyses we inspected modification indices which suggested that correlating residuals would improve model fit. A justification could be made for correlating residuals as all items began with the stem: "Confio en mi que…" ("I have confidence in…."). However, correlating residuals is not considered best standard practice so we did not proceed with testing a model with correlated residuals.

Limitations and Implications for Future Research
Our study has several limitations including a potential threat to external validity as our sample only includes noninjection crack users. Consequently, findings from this study may only generalize to individuals who use crack/cocaine and not to injection drug users. Furthermore, we employed a mono-method as most of the data was collected via self-report. Lastly, future researchers could identify additional items for the condom use selfefficacy measure, especially as one of the factors is a two-item structure. A two-item factor tends to be unstable and ill-defined thus leading to potential methodological issues.

Conclusion
Our results provide evidence that people who use substances can provide reliable answers. Before the present study most measures that have been validated on PWUDs have been measures that assess drug use [61]. Our validated measures are among the first measures validated on PWUDs that are not about drug use which provide further evidence that PWUDs can reliably self-report. Our study represents a significant contribution to the fields of infectious disease and substance use as they fill an important gap in current research, namely the lack of measures that have been validated on both Spanish-speaking individuals and people who use substances. The Medical College of Wisconsin IRB approved the study on April 18, 2011. All participants invited to participate in the study received information about the purpose of the study, that participation was voluntary, study procedures (e.g., nature of questions being asked, approximate time to completion), contact information, and potential risks and benefits of participating prior to consent. Participants provided consent by signing and dating the informed consent sheet.

Consent for Publication
Not Applicable

Availability of Data and Material
The datasets used and/or analysed during the current study are available from the corresponding author on